🎨 Trained for use with NoobAi V-Pred 1.0LyCORIS trained on a dataset of my own art from 2024!This model was trained using a collection of my own illustrations, mostly from 2024. A portion of the dataset includes images sourced from Danbooru with accompanying tags, but since I stopped posting art publicly around September 2024, I manually tagged the more recent pieces myself. I took care to curate the dataset thoughtfully to keep the overall style as consistent as possible.It is honestly very strange to think 3-4 years ago I had a deep dislike for AI. Now It has genuinely changed my life and I have made many friends because of it and I can't imagine my life without it.I don’t share my work on social media anymore, so if you happen to recognize this style, I kindly ask that you don’t mention or share my previous usernames or accounts. Thank you for understanding!🧪 Trained locally on Kohya_ss using a curated dataset of 26 images 💡 For best results, use with checkpoints developed from NoobAi V-Pred 1.0! ⚠️ Compatibility with other models besides the base isn’t guaranteed!<hr />⚠️ Non-Commercial Use Only These models are made with love for the community! Please use them for personal and non-commercial purposes only. No commissions, merch, or monetized content.<h3 id="-gifktac49">📄以下是原始英文说明的中文翻译，供中文用户参考</h3>🎨 使用 NoobAi V-Pred 1.0 训练完成这是一个以我自己在 2024 年创作的插画为基础训练的 LyCORIS 模型！这个模型使用的是我在 2024 年创作的大量插画作品进行训练。其中一部分数据集来源于 Danbooru，并附有标签；但由于我在 2024 年 9 月左右停止公开发布作品，因此较新的部分是由我手动标注的。我在整理数据集时非常用心，尽量保持整体画风的一致性。说实话，想到三四年前我还非常反感 AI，现在却因为它彻底改变了我的生活，甚至结识了很多好朋友，真的很不可思议。我已经无法想象没有 AI 的生活了。我现在已经不再在社交媒体上分享我的作品了，所以如果你认出了这种画风，恳请你不要提及或传播我以前使用的用户名或账号。感谢你的理解！🧪 使用 Kohya_ss 在本地训练，共使用了 105 张精选图像的数据集 💡 为获得最佳效果，建议配合由 NoobAi V-Pred 1.0 开发的模型使用！ ⚠️ 除了该基础模型以外，与其他模型的兼容性不予保证！<hr />⚠️ 仅限非商业用途 这些模型是出于热爱分享给社区的！请仅将其用于个人、非商业目的。请勿用于稿件、周边、或其他任何营利性内容。

[NoobAI] Artist Style - Cococolacake

This version incorporates a slightly larger and refined dataset. I was planning to retrain this for a second version in the future, using some updated settings, but I am actually quite satisfied with this version. If you find the lora is too strong, lower the weight, the network alpha = dim during training so it can sometimes be too strong.For some reason, this file seemed to be deleted. I am not sure why, so I have reuploaded it.

artist_cococolacake_vpred_v1-1.safetensors

2024 v1.1 VPRED

v1.1 VPRED

2024 v1.0 VPRED

After looking through training I realized some images in the dataset were very low resolution due to a mistake on my end, and had transparent backgrounds. I retrained with a slower LR and more epoch after fixing this, it resulted in much more stable images but looks slightly different.Obviously it is up to personal preference which is used.

Files with SHA256: b86163fb24704828ee215c77401820c5746da4c5735b098f1fe0cae4344b49ea